SDTM to ADaM — ADSL, BDS, and OCCDS
Synthetic data. Not submission-ready. Not a copy of official IG mapping tables. Licensed vocabularies are not republished here.
The problem
Analysis datasets should be traceable back to tabulation data. Official CDISC ADaM materials describe analysis-ready structures and the idea that ADaM should make the path from SDTM to analysis clear. Open-source examples of implementing those ideas in R live in pharmaverse and admiral. This page names the common structures a data engineer will hear, and one synthetic derivation — it is not an ADaM IG.
A working example
Official and pharmaverse materials commonly talk about:
- ADSL — one row per subject, with population flags and key subject-level dates/treatments. Other ADaM datasets typically need ADSL.
- BDS (Basic Data Structure) — one or more rows per subject per parameter, often with a result, baseline, and change-from-baseline. Findings (labs, vitals) often land here.
- OCCDS (occurrence structures) — event-style analysis rows (adverse events, prior medications, and similar).
Illustrative flow on the synthetic subjects (not executed as a full ADaM library, just the logic):
- Start from synthetic DM-like rows (SYN-001, SYN-002) to build a subject-level ADSL-like table: one row each, with a study identifier and any treatment-start date your source actually has. If you do not have a treatment start, do not invent one.
- For a vitals finding (SYN-001 heart-rate-local-code = 72 at 2026-02-14T08:00:00), a BDS-like row would carry the subject key from ADSL, a parameter, the result, and — only if a baseline definition is documented — a baseline flag or change. This site has no protocol, so no baseline is derived.
- For AE-like rows, an OCCDS-like table would keep the verbatim, the timing, and any coded term produced under a licensed process. It should still join to ADSL on the subject key.
Pharmaverse’s documented workflow is: SDTM → ADSL → other ADaM → tables/listings/figures. Official admiral material describes an ADaM dataset as a sequence of derivations (functions that add variables or records). That is a good engineering habit even outside R: each derived column should have a named, testable rule.
The gotcha
ADSL is not optional plumbing. If you build an analysis lab table first and join demographics ad hoc, two findings will eventually disagree about a population flag.
Change from baseline is a definition, not a formula you can guess. Which visit is baseline, how missing baselines are handled, and whether you derive them in ADaM or receive them from SDTM is a statistical/mapping-spec question.
Traceability. Official ADaM emphasis on traceability means a reviewer should get back to the SDTM row(s) that fed a derived value. See traceability.
When not to use this
- Do not produce “ADaM-like” datasets for a filing from this page.
- Do not port admiral function names into SQL and assume conformance.
- Do not derive efficacy endpoints without the SAP / analysis specs.
Structure names and the SDTM → ADSL → other ADaM flow are described in public CDISC and pharmaverse / admiral documentation linked below. No official ADaM IG text is reproduced. The synthetic vitals and AE rows are the same DuckDB examples used elsewhere on this site. No submission-ready ADaM was generated.